Information fusion

1
Information fusion

• Steve Roberts
• Machine Learning Research Group
• University of Oxford

2
Information fusion

• Reduction in uncertainty about variable of
  interest but what is of interest
• Fusion higher level representation or
  abstraction
• Model of plausible relationships between
  variables problem of causality
• Inference model building in the presence of
  uncertainty

3
Fusion as learning causation
?
4
Direct causation
5
Latent variable or explanation
6
Multiple latent variables
7
Recipe for fusion

• Handling uncertainty
• Inferring plausible model
• Observed data streams do not represent the state
  of interest directly
• THINK GENERATIVE
• Probabilistic reasoning
• Dynamic inference
• Hypothesis evaluation

8
In summary
FUSION DOES NOT ACT FROM OBSERVATIONS TO STATE OF
INTEREST
S.O.I.
X
Y
9
In summary
FUSION DOES ACT GENERATIVELY
S.O.I.
X
Y